CN115482912A - Self-help psychological intervention system and method for conversation machine - Google Patents

Abstract

The invention discloses a self-help psychological intervention system of a conversation machine, which comprises: a psychological dialogue exchange terminal for interaction with a user; an extreme emotion recognition system for recognizing an extreme emotion; and the consultation conversation fusion device is used for selecting and outputting contents filtered and output by the subsystems according to a preset priority order, and outputting the contents filtered and output by the psychological professional conversation recommendation system when the extreme emotion recognition system recognizes extreme emotion. The invention has the beneficial effects of identifying extreme emotion and achieving self-help psychological intervention. The invention provides a self-service psychological intervention method for a conversation machine, which comprises the following steps: recognizing input text of a user; identifying whether the input text belongs to one or more preset emotion models, and if so, recommending and outputting a matched psychology specialty; and if not, filtering and recommending psychological professional content or psychological scene response data or chatting response data. The method has the beneficial effects of identifying extreme emotion and achieving self-help psychological intervention.

Description

Self-help psychological intervention system and method for conversation machine

Technical Field

The invention relates to the technical field of artificial intelligence. More particularly, the invention relates to a self-help psychological intervention system and a self-help psychological intervention method for a conversation machine.

Background

The basic idea of psychological service is to help others to help themselves, namely, a person with psychological problems learns a set of psychological defense mechanism through the guidance of a consultant, and the person can help himself when the next psychological problem appears. However, there are currently two difficulties with this approach. First, the resources of the consultant are limited and expensive, resulting in many psychological patients being delayed or unable to pay for the consultation. On the other hand, many people with psychological problems are reluctant to actively search for counseling services for reasons of pubic affection, thereby missing the opportunity for intervention.

In the face of the difficulty of manual psychological consultation, a plurality of products based on psychological contents such as psychological assessment, popular science articles, psychological videos, meditation audios and the like are born, and a self-service is hoped to be provided for users. However, these products expect a particularly strong motivation for users to acquire and use such content, which many users do not know or use completely because of the lack of initiative.

The conversation robot technology is delivered by a man-machine conversation mode and a user, so that the user can be stimulated to express more ideas, the difficulty and the cost of obtaining psychological self-service by the user are reduced, and the psychological self-service is accepted more and more. However, there are two contradictions that are difficult to reconcile in the psychological service and intervention that dialog robot systems currently exist on the market. First, many conversational robotic systems are based on chatting and other living needs, do not have professional knowledge of psychological services, and do not achieve the purpose of psychological intervention. Secondly, some dialog systems based on mental expert knowledge have strong specialization, but the user looks boring and uninteresting in the using process, and the purpose of actively attracting the user to express the internal idea cannot be achieved.

Disclosure of Invention

An object of the present invention is to solve at least the above problems and to provide at least the advantages described later.

To achieve these objects and other advantages in accordance with the purpose of the invention, there is provided a conversation machine self-service psychological intervention system, comprising:

the psychological dialogue exchange terminal is used for interacting with the user and receiving input information of the user;

the extreme emotion recognition system is connected with the psychological conversation exchange terminal and is used for recognizing and judging the extreme emotion of the input information;

and the consulting dialogue fusion device is connected with the psychological dialogue exchange terminal and is used for selecting and outputting contents filtered and output by a plurality of subsystems to the psychological dialogue exchange terminal according to a preset priority sequence, wherein the subsystems comprise a psychological professional dialogue recommendation system, a psychological professional content recommendation system, a psychological QA big data intelligent system and a chatting big data intelligent system, the priorities of which are sequentially reduced, and when the extreme emotion recognition system recognizes extreme emotion, the consulting dialogue fusion device outputs the contents filtered by the psychological professional dialogue recommendation system to the psychological dialogue exchange terminal.

Preferably, the method further comprises: the emotion model library comprises a plurality of emotion models, wherein each emotion model comprises an emotion model threshold value, a plurality of emotion keywords and a weight coefficient of each emotion keyword;

the extreme emotion recognition system takes a text input by a user of the psychological dialogue exchange terminal as an input text, calls a word segmentation tool to process to obtain keywords of the input text, traverses each emotion model, calls an emotion word recognition algorithm to recognize emotion keywords in the input text and weights corresponding to the emotion keywords, calls an emotion recognition comprehensive scoring algorithm to calculate emotion scores of each emotion model of each input text, compares the emotion scores with corresponding emotion model thresholds, and outputs the emotion models, the recognized emotion keywords and the obtained emotion scores which are higher than the emotion model thresholds.

Preferably, the word segmentation tool segments the input text, acquires position information of the segmented words in the input text, deletes pause words and corresponding position information in the input text, identifies negative words and corresponding position information in the input text, identifies turning words and corresponding position information in the input text, identifies degree words and corresponding position information in the input text, and determines the rest words as keywords.

Preferably, the emotion word recognition algorithm includes the steps of:

a. acquiring a keyword kw1 of an input text;

b. obtaining an emotion keyword kw2 and a corresponding coefficient in an emotion model;

c. calling a Sennte-BERT model to calculate vector values v1 and v2 of the keywords kw1 and the emotion keywords kw 2;

d. calculating cosine similarity s12 of the vector v1 and the vector value v2;

e. converting the cosine similarity s12 into model similarity by adopting a formula s1= s12 × s 2;

f. judging whether a degree word is recognized in the range of the first 3 words of the keyword v1, if the degree word exists, searching the level of the degree word in a degree word dictionary, acquiring the degree weight Wj corresponding to the degree word, and updating the model similarity by adopting the degree word weight s1= s1 × Wj, wherein the degree word dictionary comprises a plurality of degree words and the degree weight of each degree word;

if no degree word exists, the model similarity s1 does not need to be updated;

g. judging whether the similarity s1 of the models reaches an emotion model threshold value, if not, repeating the steps a-f to continuously judge the next keyword;

if the emotion model threshold is reached, adding the keyword v1 and the model similarity s1 to an emotion word list, and repeating the steps a-f to continuously judge the next keyword;

h. and repeating the steps a to g until all the keywords in the input text are traversed.

Preferably, the emotion recognition comprehensive scoring algorithm comprises the following steps:

i, acquiring emotion words and model scores (model similarity s 1);

II, acquiring position information of the emotion words in the input text;

III, counting the number N of turning words in the input text;

if the number N of the turning words is an even number, a weighting coefficient W is given as 1, and if the number N of the turning words is an odd number, the weighting coefficient W is given as-1;

IV, judging whether a negative word is recognized in the range of the first 3 characters of the emotion word, and if no negative word exists, updating the comprehensive emotion score S = S + S1W;

if there is a negative word, updating the weight coefficient W = W (-1), and updating the integrated emotion score S = S + S1W;

step V, repeating the steps I to IV until all the emotion words in the emotion word list are traversed, and calculating to obtain the total score of the comprehensive emotion scores of all the emotion words, namely the emotion score;

and VI, judging whether the emotion scores reach the corresponding emotion model threshold value, and if so, outputting the corresponding emotion models, emotion words and emotion scores.

Preferably, the psychological professional tactical recommendation system is connected with a plurality of tactical libraries, each tactical library comprises a crisis intervention tactical library and a psychological intervention tactical library, and the crisis intervention tactical library has higher priority than the psychological intervention tactical library;

and when the extreme emotion recognition system recognizes that the input text belongs to one or more emotion models, starting the psychological professional jargon recommendation system, and recommending jargon of the crisis intervention jargon library or the psychological intervention jargon library.

Preferably, the system also comprises a psychological professional content library which comprises a self-psychological assessment library with preset weight, a professional memorial meditation library, a psychological science pragman library and a psychological video library;

when the extreme emotion recognition system recognizes that the input text does not belong to any emotion model, the psychology specialty content recommendation system matches the input text with psychology specialty content by a keyword matching method, then calculates the language similarity between the input text and the psychology specialty content meeting the keyword matching standard by a BERT vector similarity calculation method, updates the weight of the corresponding psychology specialty content library to be superposed on the language similarity, filters out the specialty content of the psychology specialty content library meeting the similarity threshold according to a preset similarity threshold, takes out a plurality of specialty content with the highest similarity after filtering, and randomly extracts one specialty content to be output through the consultation dialogue fusion device.

Preferably, the psychological QA big data intelligent system collects question-answer pair data of a mass of psychological consultation scenes, calls an automatic question-answer algorithm to output answer information, and outputs the answer information through the consultation dialogue fusion device.

Preferably, the big chatting data intelligent system calls a chatting question-answer interface to output the answering information, and the answering information is output through the consultation dialogue fusion device.

The method for self-help psychological intervention of the dialogue machine comprises the following steps:

step one, recognizing an input text of a user;

step two, identifying and judging whether the input text belongs to one or more preset emotion models, and if so, recommending and outputting a matched psychology specialty dialect;

if not, performing keyword matching and similarity calculation on the input text and the psychological professional contents, filtering and taking out a plurality of psychological professional contents which reach a preset similarity threshold and have the highest similarity, and randomly extracting one psychological professional content for outputting;

step three, screening answer data matched with the input text from the question-answer data of the massive psychological consultation scene when no psychological professional content reaching the similarity threshold exists;

and step four, when no matched answer data exists, selecting chatting answer data with the optimal relevance with the input text and outputting the chatting answer data.

The invention at least comprises the following beneficial effects:

firstly, the invention brings a plurality of electronic psychological intervention schemes, psychological professional contents, referral and crisis intervention information which can serve self-help psychological service to a psychological user in a personalized way through the conversation robot technology, thereby facilitating the psychological consultation of the client suffering from mild and moderate psychology to be expensive and the disorder of affection and pubic feeling, and greatly reducing the cost and the efficiency of psychological health prevention.

Secondly, the invention takes the extreme emotion recognition of the user as a key point, not only can help the user to deal with some crisis conditions in life, but also can provide good crisis intervention information when the unit uses the system.

Thirdly, the patent adapts to the demand of global psychological crisis after epidemic situation, can greatly improve the coverage of psychological service, and realizes real popularization of psychological service.

Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.

Drawings

Fig. 1 is a schematic diagram of a framework of the psychological intervention system according to one embodiment of the invention;

fig. 2 is a flow chart of the extreme emotion recognition system of one of the technical solutions of the present invention;

FIG. 3 is a category of the extreme mood model for one of the solutions of the present invention;

FIG. 4 is an example of the extreme emotion model keyword dictionary in one embodiment of the present invention;

FIG. 5 is a flowchart of an emotional word recognition algorithm according to one embodiment of the present invention;

FIG. 6 is a flow chart of a sentiment recognition comprehensive scoring algorithm according to one embodiment of the present invention;

fig. 7 is a flow chart of a man-machine interaction of the psychological intervention system according to one of the solutions of the present invention.

Detailed Description

The present invention is further described in detail below with reference to the attached drawings so that those skilled in the art can implement the invention by referring to the description text.

It should be noted that in the description of the present invention, the terms indicating orientation or positional relationship are based on the orientation or positional relationship shown in the drawings only for the convenience of description and simplification of the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore should not be construed as limiting the present invention.

As shown in fig. 1 to 7, the present invention provides a dialogue machine self-help psychological intervention system, which includes a psychological dialogue exchange terminal 101 and a consulting dialogue fuser 100, and an extreme emotion recognition system 102, a psychological professional recommendation system 103, a psychological professional content recommendation system 104, a psychological QA big data intelligent system 105, and a chatting big data intelligent system 106 connected to the consulting dialogue fuser 100. The psychological professional jargon recommendation system 103 is connected with jargon libraries such as a crisis intervention jargon library 107 and a psychological intervention jargon library 108. The psychology specialty content recommendation system 104 is connected with a psychology content library such as a psychology assessment library 109, a professional memorial meditation library 110, a psychology prague library 111, and a psychology video library 112. The mental QA big data intelligent system 105 matches a mental question with a corresponding answer by calling a deep learning model of natural language. The chatting big data intelligent system 106 answers the question posed by the user by calling the chatting database of the public platform listening.

Extreme emotion recognition system 102:

the extreme emotion recognition system 102 uses a segment of characters input by a user as input information, and finally outputs a group of extreme emotions (belonging to which emotion model), corresponding emotion scores and words and sentences (emotion words) related to the emotion through a series of calculations for continuous use in the steps of calling the algorithm.

As shown in fig. 2, the algorithm flow of the extreme emotion recognition system 102 includes the following steps:

and step 200, starting extreme emotion recognition.

Step 201, calling a Chinese word segmentation tool such as Jieba, segmenting words of a text input by a user, and acquiring position information of the segmented words in the text.

And step 202, removing the stop words and the corresponding position information in the text.

And step 203, identifying negative words and corresponding position information in the text.

And step 204, identifying the turning words and the corresponding position information in the text.

And step 205, identifying the degree words and the corresponding position information in the text.

And step 206, judging whether all emotion keyword models are traversed or not. If the traversal is finished, jumping to step 212; otherwise, step 207 is entered. This patent covers 16 different extreme emotions (emotional models) involving teenagers and young people, as shown in fig. 3, including suicide, cyber gambling, depression, cyber addiction, campus overlord, loss of love, insomnia, anxiety, sex loving, sexual assault, violence, sex addiction, drug addiction, overeating, self-mutilation, etc. Each extreme emotion is provided with a plurality of keywords and each keyword is accompanied by a coefficient relating to that emotion as shown in figure 4 which lists some of the contents of the extreme emotion model dictionary in suicide, net gambling, depression, net addiction, campus tyrant etc. 5.

And step 207, extracting the keywords and the threshold parameters of the model.

And step 208, calling an emotion word recognition algorithm to acquire emotion words related to the emotion. This step is followed by further steps, which will be described in more detail below.

And step 209, invoking an emotion recognition comprehensive scoring algorithm to calculate the emotion scores. This step is followed by further steps, which will be described in more detail below.

Step 210, judging whether the emotion score calculated in step 209 is higher than the threshold value of the emotion model taken out in step 207. If the sentiment score is higher than the threshold, go to the next step 211; otherwise, it indicates that the text does not include the emotion model, and the flow goes to step 206 to determine whether to traverse the next emotion model.

And step 211, adding parameters such as the emotion model number, the recognized keywords, the emotion scores and the like to an early warning list. The flow then branches to step 206 to determine whether to traverse the next emotion model.

And step 212, returning the identified extreme emotion early warning list. Flow goes to here that the algorithm has completed the recognition of all extreme emotions.

And step 213, ending extreme emotion recognition.

Emotion word recognition algorithm 208:

the emotion word recognition algorithm 208 calculates word-to-word and phrase-to-phrase similarities using a deep learning model, and selects related keywords and phrases by comparing the similarities to a model threshold.

The algorithm flow of emotion word recognition algorithm 208 includes the following steps:

and step 300, emotion word recognition is started.

Step 301, judging whether all the keywords in the text are traversed. If the traversal is finished, jumping to step 313 and returning to the emotion word list; otherwise, the next step is performed.

Step 302, acquiring a keyword kw1 in the text.

And step 303, judging whether all the keywords in the emotion model are traversed. If the traversal is finished, jumping to step 301, and judging whether a keyword needs to be processed; otherwise, the next step is performed.

And step 304, obtaining a current keyword kw2 and a correlation coefficient s2 in the emotion model.

And step 305, calling a sequence-BERT model to calculate vector values v1 and v2 of kw1 and kw 2. The Bert model has exhibited a strong posture in NLP's large tasks. The semantic similarity calculation (semantic similarity) task is not an exception, but due to the provision of the bert model, when calculating the semantic similarity, two sentences need to enter the model at the same time for information interaction, which causes a large amount of calculation overhead. For example, there are 10000 sentences, and we want to find the most similar sentence pair, which needs to be calculated (10000 × 9999/2) times, and takes about 65 hours. The sequence-BERT network architecture is used to address the deficiencies of the BERT model. In a simple and popular way, a frame of a twin network model is used for reference, different sentences are input into two bert models (but the two bert models are parameter-shared and can be understood as the same bert model), and a sentence characterization vector of each sentence is obtained; and the finally obtained sentence representation vector can be used for semantic similarity calculation and also can be used for an unsupervised clustering task. For the same 10000 sentences, we want to find the most similar sentence pairs, and only need to calculate 10000 times, which takes about 5 seconds to calculate the completeness.

And step 306, calculating cosine similarity s12 of the vectors v1 and v 2.Cosine similarity (Cosine similarity) is a normalized calculation method, and text similarity is suitable to be calculated.

Step 307, converting the vector cosine similarity s12 into a model similarity s1= s12 × s2.

And 308, judging that the recognized degree words exist in the range of the first 3 characters of the emotion words. A degree word dictionary with different levels is established in advance and can be quickly consulted. If the range of the 3 characters has degree words, the next step 309 is carried out; otherwise, go to step 311.

Step 309, find out the degree word level, and obtain the corresponding degree weight Wj.

And step 310, updating the model similarity s1= s1 × Wj by using the degree word weight Wj.

And 311, judging whether the model similarity s1 reaches a model threshold value. If not, skipping to step 303 to process the next model keyword; otherwise, the next step 312 is entered.

And step 312, adding the keyword kw1 and the similarity s1 to the emotional word list. Then, the process jumps to step 303 to process the next model keyword.

And step 313, returning to the emotion word list.

And step 314, finishing emotion word recognition.

Emotion recognition comprehensive scoring algorithm 209:

the emotion recognition comprehensive scoring algorithm 209 calculates a comprehensive emotion score of the text by considering the negative words and the turning words for all the emotion words recognized by the emotion word recognition algorithm 208. Our algorithm can be simplified to some extent because the whole algorithm focuses on extreme emotions, not the whole emotion.

The algorithm flow of the emotion recognition comprehensive scoring algorithm 209 includes the following steps:

and step 400, starting comprehensive scoring of the emotion.

Step 401, initializing a score S =0 for one extreme emotion model.

And step 402, judging whether all the recognized emotion words are traversed. If the traversal is finished, jumping to a step 412 and returning the total score; otherwise, the next step 403 is entered.

And step 403, acquiring the current emotion word kw1 and the corresponding model score s1.

And step 404, acquiring the position information of the current emotion word in the text.

And step 405, counting the number of turning words in the text, and setting the number as num (N).

And step 406, judging whether the number num of the turning words is an even number. If the number is even, go to step 407; otherwise, i.e. odd, step 408 is skipped.

Step 407, a weighting factor W is given as 1.

Step 408, the weighting factor W is given as-1.

And step 409, judging whether a negative word is recognized in the range of the first 3 words of the emotional word. If no negative word, jump to step 411; otherwise, the next step 410 is entered.

And step 410, updating the weight coefficient W = W (-1).

And step 411, updating the comprehensive emotion score S = S + S1W.

And step 412, returning the total score S.

The professional jargon recommendation system 104:

the mental professional jargon recommendation system 103 calls the jargon of the crisis intervention jargon library 107 and the mental intervention jargon library 108 in turn according to a set priority, and determines which jargon is enabled or not enabled according to a set threshold. The psychological expert speech recommendation system 103 builds a language model for each of the crisis intervention speech library 107 and the psychological intervention speech library 108, allowing one speech to correspond to a plurality of natural language expressions. The psychological expert tactics recommendation system 103 has different calculation models for the crisis intervention tactics library 107 and the psychological intervention tactics library 108. For the crisis intervention tactics library 107, the psychology specialty tactics recommendation system 103 uses a set of emotion analysis computing programs to determine the tendency of the user language to any extreme emotion. If the trend value is greater than a set threshold, the corresponding extreme emotional technique is initiated. For the psycho-interventional speech library 108, the psycho-professional speech recommendation system 103 uses a set of similarity calculation programs based on BERT vectors to calculate which of the professional languages the user language has the highest similarity to. If the highest similarity is greater than a set threshold, the corresponding professional routine is initiated. Because critical interventions are more important than general psychological conversational interventions, the psychological conversational recommendation system 103 considers first recommendations for critical interventions and then recommendations for professional psychological conversational interventions.

The crisis intervention dialect library 107 and the psychological intervention dialect library 108 adopt a current dialogue robot technical dialogue form, one dialect is equivalent to a computer program of one man-machine dialogue, and consists of a plurality of computer prompt questions and a plurality of answers. The same question, depending on the manual answer, the computer may respond with different answers, guide the utterance and some specialized psychological tools or further guide the user. The crisis intervention dialoging library 107 manages a series of dialogue procedures for self-crisis intervention. These dialog programs diagnose crisis, mediate crisis intervention enthusiasm, and invoke psycho-and emotional conditioning for intervention. The crisis intervention tactics library 107 comprises 16 extreme emotions such as suicide, net gambling, depression, net addiction, campus overlord, loss of love, insomnia, anxiety, sex loving, sexual assault, household violence, sex addiction, drug addiction, overeating, self-disabled, anorexia and the like. The psychological intervention dialogs library 108 manages a series of CBT-based dialogs. CBT is based on the idea that our interpretation of events affects our thought and feel. By learning to think about the problem in different ways, to re-conceive, to change certain behaviors, to make the user feel better and more able to live through his own intended life. Since 2014, electronic cognitive behavioral therapy (CCBT) has developed vigorously, and more clinical trials have demonstrated its effectiveness, even achieving more effective and popular effects than conventional face-to-face treatment approaches. The mental intervention words library 108 includes "relieving mental anxiety conversation", "relieving depressed mood conversation", "improving insomnia conversation", "walking down mood conversation", "relieving autism conversation", "relieving anger mood conversation", "reducing mental sad conversation", "relieving depressed mood conversation", "coping with apathy mood conversation", etc.

Psychology specialty content recommendation system:

the psychology professional contents recommendation system 104 considers the professional psychology contents from the psychological assessment library 109, the professional meditation library 110, the psychology prague library 111 and the psychology video library 112 at the same time according to different weights, and randomly selects the professional psychology contents from the best matching contents to recommend to the user. The psychology specialty content recommendation system 104 builds a language model for each of the psychology evaluation library 109, the professional meditation library 1107, the psychology prague library 111, and the psychology video library 112, allowing one content to correspond to multiple natural language expressions. The psycho-professional content recommendation system 104 calculates the language similarity between the user language and professional content meeting the keyword matching criteria by using a set of similarity calculation programs based on BERT vectors based on a conventional keyword matching result, and superimposes the weights of different content libraries on the similarity. And filtering professional contents reaching the threshold value according to a preset similarity threshold value, taking out a plurality of professional contents with the highest similarity after filtering, and randomly extracting contents from the professional contents and recommending the contents to the user.

QA big data intelligent system 105:

the mental QA big data intelligent system 105 collects and manages massive question and answer pair data based on a mental consultation scene, considers the knowledge map information of the mental consultation and calls an automatic question and answer algorithm to output answers.

Chat with big data intelligent system 106:

the chatting big data intelligent system 106 acquires answers by calling a relatively mature chatting question-answer interface in the market at present, and the answers may not have the specificity of psychological consultation but play a good connection role in guaranteeing the fluency of conversation.

The consultation session fuser 100:

the consulting dialogue fuser 100 calls the contents output by the subsystems such as the extreme emotion recognition system 102, the psychology professional skill recommendation system 103, the psychology professional content recommendation system 104, the psychology QA big data intelligent system 105 and the chatting big data intelligent system 106 according to a set priority in turn, and determines which subsystem is output as the output of the psychological dialogue exchange terminal 102 according to a set filtering principle.

Referring to fig. 2, the consultation conversation fusion apparatus 100 constructs a human-machine conversation process using the above system, including the following steps:

and step 0, starting any round of mental conversation.

Step 1, acquiring the text content input by a user.

And 2, identifying extreme emotions from the text.

And 3, judging whether extreme emotions exist in the text. If the extreme emotions exist, entering a step 4, and entering an identified extreme emotion intervention module; if there is no extreme emotion, step 5 is entered and an attempt is made to recommend professional psychological intervention.

And 4, entering the identified extreme emotion intervention module.

And 5, recommending professional psychological intervention.

And 6, judging whether psychological and operative intervention is needed. If appropriate professional psychological intervention exists, entering a step 8, and entering a recommended psychological intervention dialogue module; if no appropriate intervention is available, step 7 is entered, attempting to recommend the psychological content.

And 7, recommending professional psychological contents.

And 8, entering a recommended psychological intervention dialogue module.

And 9, judging whether appropriate professional psychological content recommendation exists or not. If appropriate professional psychological content recommendation exists, entering a step 10, and pushing appropriate professional psychological content for the user to use; if no professional psychological content recommendation is available, step 11 is entered, and the psychological question-answering system is invoked.

And 10, pushing out proper professional psychological contents for the user to use.

And step 11, calling a psychological question-answering system.

And step 12, judging whether appropriate psychological question and answer content recommendation exists. If the appropriate psychological question and answer content recommendation exists, entering a step 13, and outputting a question and answer; if there is no appropriate psychological quiz content recommendation, step 14 is entered and a chat conversation interface is invoked.

And step 13, outputting the question answer.

And step 14, calling a plurality of chatting dialogue interfaces.

And step 15, evaluating and selecting the chatting content. And performing comprehensive evaluation on the chatting answers from the multiple interfaces by considering the relevance of the chatting answers to the questions and the answer quality, and selecting the most appropriate chatting conversation answer.

And step 16, judging whether a proper chatting dialogue answer exists or not. If the chat answers are correct, entering step 17, and outputting the chat answers; if there is not a proper chat answer, step 18 is entered and no text is output.

And step 17, outputting the chatting content.

And step 18, outputting the text which is not answered.

And step 19, ending the psychological dialogue in the current round. The conversation process is actually hung there, and the step 0 can be returned to continue the conversation.

The invention discloses a self-service psychological service robot system for identifying various extreme emotions and fusing various data and knowledge, which adopts a standard B/S framework, manages professional psychological contents, particularly conversation logic, by a data background used by a psychological consultation professional team, recommends the most appropriate psychological intervention contents to a user by a psychological user through analyzing the intention of the user by various artificial intelligence algorithms, and achieves the purpose of self-service acquisition of psychological intervention by the psychological user. The system has two bright spots, one of the bright spots is the extreme emotion recognition algorithm with high accuracy and wide range, so that not only can important clues be provided for a mental health manager, but also the accuracy of a recommendation algorithm for extreme emotion users can be improved; secondly, the recommendation algorithm of the system integrates various forms of contents including professional dialogue consultation, professional psychological assessment, professional memorial meditation exercise, psychology common scrip and videos, psychological question answering, chatty dialogue and the like, so that the content in dialogue delivery is greatly enriched, and a large space is provided for continuously optimizing personalized intervention effects.

While embodiments of the invention have been described above, it is not limited to the applications set forth in the description and the embodiments, which are fully applicable in various fields of endeavor to which the invention pertains, and further modifications may readily be made by those skilled in the art, it being understood that the invention is not limited to the details shown and described herein without departing from the general concept defined by the appended claims and their equivalents.

Claims (10)

1. Dialogue machine self-service psychological intervention system, characterized by comprising:

the psychological dialogue exchange terminal is used for interacting with the user and receiving input information of the user;

the extreme emotion recognition system is connected with the psychological conversation exchange terminal and is used for recognizing and judging the extreme emotion of the input information;

and the consultation conversation fusion device is connected with the psychological conversation exchange terminal and is used for selecting and outputting contents which are filtered and output by a plurality of subsystems to the psychological conversation exchange terminal according to a preset priority sequence, wherein the subsystems comprise a psychological professional conversation recommendation system, a psychological professional content recommendation system, a psychological QA big data intelligent system and a chatting big data intelligent system, the priorities of which are sequentially reduced, and when the extreme emotion recognition system recognizes extreme emotion, the consultation conversation fusion device outputs the contents filtered by the psychological professional conversation recommendation system to the psychological conversation exchange terminal.

2. The conversational machine self-service psychological intervention system of claim 1, further comprising: the emotion model library comprises a plurality of emotion models, wherein each emotion model comprises an emotion model threshold value, a plurality of emotion keywords and a weight coefficient of each emotion keyword;

the extreme emotion recognition system takes a text input by a user of the psychological conversation exchange terminal as an input text, calls a word segmentation tool to process to obtain keywords of the input text, traverses each emotion model, calls an emotion word recognition algorithm to recognize the emotion keywords in the input text and weights corresponding to the emotion keywords, calls an emotion recognition comprehensive scoring algorithm to calculate emotion scores of the emotion models of each input text, compares the emotion scores with corresponding emotion model thresholds, and outputs the emotion models, the recognized emotion keywords and the obtained emotion scores which are higher than the emotion model thresholds.

3. The system of claim 2, wherein the segmentation tool performs segmentation on the input text, and simultaneously obtains position information of the segmentation in the input text, deletes a stop word and corresponding position information in the input text, identifies a negative word and corresponding position information in the input text, identifies a turning word and corresponding position information in the input text, identifies a degree word and corresponding position information in the input text, and identifies the remaining words as keywords.

4. The conversational machine self-service psychological intervention system of claim 3, wherein the recognize emotional words algorithm comprises the steps of:

a. acquiring a keyword kw1 of an input text;

b. obtaining an emotion keyword kw2 and a corresponding coefficient in an emotion model;

c. calling a Sennce-BERT model to calculate vector values v1 and v2 of the keywords kw1 and the emotion keywords kw 2;

d. calculating cosine similarity s12 of the vector v1 and the vector value v2;

e. converting the cosine similarity s12 into model similarity by adopting a formula s1= s12 × s 2;

f. judging whether a degree word is recognized in the range of the first 3 words of the keyword v1, if so, searching the level of the degree word in a degree word dictionary, acquiring the degree weight Wj corresponding to the degree word, and updating the model similarity s1= s1 × Wj by adopting the degree word weight, wherein the degree word dictionary comprises a plurality of degree words and the degree weight of each degree word;

if no degree word exists, the model similarity s1 does not need to be updated;

g. judging whether the similarity s1 of the models reaches an emotion model threshold value, if not, repeating the steps a-f to continuously judge the next keyword;

if the emotion model threshold is reached, adding the keyword v1 and the model similarity s1 to an emotion word list, and repeating the steps a-f to continuously judge the next keyword;

h. and e, repeating the steps a to g until all the key words in the input text are traversed.

5. The conversational machine self-service psychological intervention system of claim 4, wherein the emotion recognition composite scoring algorithm comprises the steps of:

i, acquiring emotion words and model scores (model similarity s 1);

II, acquiring position information of the emotion words in the input text;

III, counting the number N of turning words in the input text;

if the number N of the turning words is an even number, a weighting coefficient W is given as 1, and if the number N of the turning words is an odd number, the weighting coefficient W is given as-1;

IV, judging whether a negative word is recognized in the range of the first 3 characters of the emotion word, and if no negative word exists, updating the comprehensive emotion score S = S + S1W;

if there is a negative word, updating the weight coefficient W = W (-1), and updating the integrated emotion score S = S + S1W;

v, repeating the steps I-IV until all the emotional words in the emotional word list are traversed, and calculating to obtain the total score of the comprehensive emotional scores of all the emotional words, namely the emotional score;

and VI, judging whether the emotion scores reach the corresponding emotion model threshold, and if so, outputting the corresponding emotion models, emotion words and emotion scores.

6. The dialogue machine self-service psychological intervention system of claim 5, wherein the psychological professional jargon recommendation system is coupled to a plurality of jargon libraries, the jargon libraries including a crisis intervention jargon library and a psychological intervention jargon library, the crisis intervention jargon library having a higher priority than the psychological intervention jargon library;

and when the extreme emotion recognition system recognizes that the input text belongs to one or more emotion models, starting the psychological professional jargon recommendation system, and recommending jargon of the crisis intervention jargon library or the psychological intervention jargon library.

7. The conversational machine self-service psychological intervention system of claim 5, further comprising a psychographic content library comprising a pre-weighted self-psychological assessment library, a professional meditation library, a psychographic script library, a psychovisual library;

when the extreme emotion recognition system recognizes that the input text does not belong to any emotion model, the psychology specialty content recommendation system matches the input text with psychology specialty content by a keyword matching method, then calculates the language similarity between the input text and the psychology specialty content meeting the keyword matching standard by a BERT vector similarity calculation method, updates the weight of the corresponding psychology specialty content library to be superposed on the language similarity, filters out the specialty content of the psychology specialty content library meeting the similarity threshold according to a preset similarity threshold, takes out a plurality of specialty content with the highest similarity after filtering, and randomly extracts one specialty content to be output through the consultation dialogue fusion device.

8. The conversational machine self-help psychological intervention system of claim 5, wherein the psychological QA big data intelligence system collects question-answer pair data of a mass of psychological consulting scenes, the psychological QA big data intelligence system calls an automatic question-answer algorithm to output answer information, and the answer information is output through the consulting conversational fuser.

9. The dialogue machine self-help psychological intervention system of claim 5, wherein the chatting big data intelligent system calls a chatting question-answer interface to output the answering information and outputs the answering information through the consultation dialogue fuser.

10. A dialogue machine self-service psychological intervention method based on the system of any one of claims 1 to 9, comprising the steps of:

step one, recognizing an input text of a user;

step two, identifying and judging whether the input text belongs to one or more preset emotion models, and if so, recommending and outputting a matched psychology specialty dialect;

if not, performing keyword matching and similarity calculation on the input text and the psychological professional contents, filtering and taking out a plurality of psychological professional contents which reach a preset similarity threshold value and have the highest similarity, and randomly extracting one psychological professional content for output;

step three, screening answer data matched with the input text from the question-answer data of the massive psychological consultation scene when no psychological professional content reaching the similarity threshold exists;

and step four, when no matched answer data exists, selecting chatting answer data with the optimal relevance with the input text and outputting the chatting answer data.

Images